Thread Rating:
  • 0 Vote(s) - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
Calculating the residues of \(\beta\); Laurent series; and Mittag-Leffler
#1
So, this is a bit off topic, but as I haven't had any quite "eureka moments" lately. I started compiling as much information as I can about \(\beta\). And it occurred to me, I do not know the Laurent series of \(\beta\) about each singularity. And as I started evaluating it, it's actually very interesting.

We're going to work solely with \(\lambda =1\) and \(b=1\) (but the result is generalizable to all \(\lambda,\beta\)) so that,

$$
\begin{align}
\beta(s) &= \Omega_{j=1}^\infty \dfrac{e^z}{1+e^{j-s}}\,\bullet z\\
\beta(s+1) &= \dfrac{e^{\beta(s)}}{1+e^{-s}}\\
\end{align}
$$

And we're looking for the laurent series of:

$$
\beta(s+j+\pi i) = \sum_{k=-\infty}^\infty a_{jk} s^k\\
$$

For \(j\ge1\). The first surprising fact, is that the laurent series exists! Second of all, I'm going to run through an induction protocol which, well, I don't know how to explain, but it's very god damned interesting.


To begin, the first pole is at \(j=1\) and it's a simple pole. This can be derived because,

$$
\begin{align}
\beta(1+\pi i + s) &= \dfrac{\exp(\beta(\pi i + s))}{1+\exp(-\pi i - s)}\\
&= \dfrac{\exp(\beta(\pi i + s))}{1-\exp(- s)}\\

\end{align}
$$

And since \(\exp(\beta(\pi i + s))\) is holomorphic about \(s = 0\); and since the pole at \(s = 0\) for \(\frac{1}{1-\exp(- s)}\) is simple. We must have this pole is simple. Additionally we have that:

$$
\frac{1}{1-\exp(- s)} = \frac{1}{s} + g(s)\\
$$

For a holomorphic function \(g(s)\) in a neighborhood of \(s = 0\). This is derived from Cauchy's formula,

$$
\begin{align}

\text{Res}_{s=0} \frac{1}{1-\exp( - s)} &= \lim_{s \to 0} \frac{s}{1-\exp(- s)}
&= \lim_{s \to 0} \frac{1}{\exp(- s)}\,\,\text{by Hopital} = 1\\
&= 1
\end{align}
$$

From here, the first singularity of \(\beta\) looks like:

$$
\beta(1+\pi i +s) = \frac{e^{\beta(\pi i + s)}}{1 - e^{- s}}\\
$$

Which must be a simple singularity. Therefore, when we look at \(\beta(1+\pi i + s)\) we get something really nice. The residue/ laurent series is very well behaved.


$$
\int_{|s| = 1/2} \beta(1+\pi i + s)\,ds = \int_{|s| = 1/2} \frac{\exp(\beta(\pi i + s))}{1-e^{- s}}\,ds
$$

Which, by Cauchy's integral theorem, we get:

$$
\int_{|s| = 1/2} \beta(1+\pi i + s)\,ds = 2 \pi i \exp(\beta(\pi i))\\
$$

From here, we can say that:

$$
\beta(1+\pi i + s) = \frac{\exp(\beta(\pi i))}{s} + h(s)\\
$$

Where \(h\) is holomorphic in a neighborhood of \(s =0\)



Now, what's so surprising is how well the higher order singularities for \(j \ge 2\) follow this pattern. And that, each Laurent series actually follows a tetration pattern! I'm going to work through \(j=2\) which is the first induction step to enlighten.

Consider:

$$
\begin{align}
\beta(2+\pi i + s) &= \frac{e^{\beta(1+\pi i + s)}}{1-e^{-1 - s}}\\
&= \frac{e^{\displaystyle e^{\beta(\pi i)}/s + h(s)}}{1-e^{-1 - s}}\\
\end{align}
$$

Now... Quite perfectly! the only singularity that arises is in the exponent as \(s \to 0\). And, although it's an essential singularity--it's the best kind of essential singularity. Good ol' fashion \(e^{1/s}\), which certainly has a laurent series.  Let's go ahead and calculate it.

$$
\beta(2 + \pi i +s) = \frac{e^{h(s)}}{1-e^{-1 -s}}\sum_{k=0}^\infty \frac{e^{\beta(\pi i) k}}{s^{k}k!}
$$

Now the real cool part happens. If we take the logarithm, we get:


$$
\log\beta(2 + \pi i +s) = \log\frac{e^{h(s)}}{1+e^{-\pi i -1 -s}} + \frac{e^{\beta(\pi i)}}{s}
$$

So, we get the recursion:

$$
\begin{align}
\log\beta(2 + \pi i +s) - \log\frac{e^{h(s)}}{1-e^{-1 -s}} &= \frac{e^{\beta(\pi i)}}{s}\\
&= \beta(1+ \pi i + s) - h(s)\\
\end{align}
$$



And now... we just apply the functional equation recursively. I'm going to write out the notation I use, which is:

$$
\beta(s + j + \pi i) = \Omega_{k=1}^{j-1} \frac{e^{z}}{1-e^{j-k-s}}\bullet \,e^{\beta(\pi i)}/s + h(s)\,\bullet z\\
$$

Which means, if:

$$
\begin{align}
a_j(s,z) &= \Omega_{k=1}^{j-1} \frac{e^{z}}{1-e^{j-k-s}}\bullet z\\
&= q_1(s,q_2(s,...,q_{j-1}(s,z)))\\
q_k(s,z) &= \frac{e^{z}}{1-e^{j-k-s}}\\
\end{align}
$$

Then:

$$
\beta(s + j + \pi i) = a_j(s,e^{\beta(\pi i)}/s + h(s))\\
$$

This is just a standard iteration of the functional equation. Now, \(a_j(s,z)\) is holomorphic in z, and in a neighborhood of zero in s. This makes the Laurent series findable; but even better a nice tetration like structure to the singularity.

Which is, in a simple terms. The value of \(\beta(s+j+\pi i) \approx \exp^{j-1} \left( e^{\beta(\pi i)}/s + h(s)\right)\). Or that, we should expect:

$$
\log^{j-1} \beta(s+j + \pi i) = \frac{e^{\beta(\pi i)}}{s} + h_j(s)\\
$$

Which means we should have a very regular structure to the singularities; and they look a lot like tetration.

Isn't that cool!

And now we can continue and find a laurent series in the neighborhood of each singularity... but that's a bit of a pain since we have a closed form expression.

Now... Onto seeing if we can find a Mittag-Leffler expansion for \(\beta\) that is valid in all of \(\mathbb{C}\). I always thought this would be impossible because the singularities would be such a mess. But with this theorem; it's probably possible. All the principle parts should look like \(f(1/(s-j-\pi i))\) for \(f\) entire. Alright!

I'll update this thread if I find anything worthy to share.



Alright, it's not quite Mittag-Leffler but it's close! Very damn close.

If you take:

$$
f_{jk}(s) = \text{Principal part of}\left(\exp^{j-1}\left(\frac{e^{\beta(\pi i)}}{s-j-(2k+1)\pi i} + h_{j}(s)\right)\right)\\
$$

This should be holomorphic on \(\mathbb{C}/\{j+(2k+1)\pi i\}\) (we've renormalized \(h_{j}\) too, just for convenience). Now the principle part, is essentially, just take the laurent series and only consider the negative exponents. So, the principle part of:

$$
\text{Principal part of}\left( \sum_{c=-\infty}^\infty a_c (z-p)^{c} \right) = \sum_{c=-\infty}^{-1} a_c (z-p)^{c}\\

$$

Now, expand the taylor series of \(f_{jk}(s)\) about \(0\). The functions \(f_{jk}\) are holomorphic about zero in increasing disks as \(j,k \to \infty\). (That's a non problem.) Take the \(2^{j+|k|}\) first terms and call this \(P_{jk}(s)\). Then,

$$
H(s) = \sum_{j=1}^\infty \sum_{k=-\infty}^\infty f_{jk}(s) - P_{jk}(s)
$$

Is a holomorphic function for \(s \neq j + (2k+1)\pi i\). This is just a slight modification of Mittag-Leffler to allow for essential singularities. Normally this doesn't work, I had to double check the literature, and the particular book I'm using: Reinhold Remmert "Classical Topics in Complex Functions Theory"--allows for the possibility of essential singularities (it's just that Mittag-Leffler's decompisition is not necessary like it is with meromorphic functions; i.e: you have to pay close care). At each singularity, it has the same principle part as \(\beta(s)\). From here, we now do something familiar to Mittag-Leffler again, and:

$$
\Pi(s) = \beta(s) - H(s)\\
$$

Which will be an entire function on \(\mathbb{C}\).

Again, this is just a sketch for the moment. But this may help... I believe we can remove the singularities of \(\beta\) and produce something like a Mittag-Leffler decomposition. This is going to take much more time though. Still not sure the best way of approach of rigorizing this. But it should look something like this.

I don't know how much this will help tetration per se. It's probably more of a curiousity with the \(\beta\) function. But you never know!
Reply


Possibly Related Threads…
Thread Author Replies Views Last Post
Question Tetration Asymptotic Series Catullus 13 429 48 minutes ago
Last Post: Catullus
Question Formula for the Taylor Series for Tetration Catullus 8 416 06/12/2022, 07:32 AM
Last Post: JmsNxn
  The beta method thesis JmsNxn 9 1,087 04/20/2022, 05:32 AM
Last Post: Ember Edison
  Trying to get Kneser from beta; the modular argument JmsNxn 2 440 03/29/2022, 06:34 AM
Last Post: JmsNxn
  tommy beta method tommy1729 0 525 12/09/2021, 11:48 PM
Last Post: tommy1729
  Trying to find a fast converging series of normalization constants; plus a recap JmsNxn 0 534 10/26/2021, 02:12 AM
Last Post: JmsNxn
  Arguments for the beta method not being Kneser's method JmsNxn 54 14,714 10/23/2021, 03:13 AM
Last Post: sheldonison
  Why the beta-method is non-zero in the upper half plane JmsNxn 0 708 09/01/2021, 01:57 AM
Last Post: JmsNxn
  Reducing beta tetration to an asymptotic series, and a pull back JmsNxn 2 1,320 07/22/2021, 03:37 AM
Last Post: JmsNxn
  Perhaps a new series for log^0.5(x) Gottfried 3 5,235 03/21/2020, 08:28 AM
Last Post: Daniel



Users browsing this thread: 1 Guest(s)